Uso de redes neurais artificiais e modelos de regressão para estimar volume de espécies nativas em Portel-PA

Abstract

Understanding the productive potential of native species contributes to sustainable management. The objective of the present study was to compare the volume estimates obtained by regression equations with artificial neural networks (RNA) for native areas under a management plan in the Portel-PA region, based on the strict cubing data of 864 trees (46 species) with DAP ≥ 45 cm of an Annual Production Units (UPA) managed in 2015. This being owned by the Uberlândia-PR farm, located in Gleba Joana Perez I s / n rural area, between the municipalities of Bagre, Portel, Baião and Oeiras in the State of Pará, in the Dense Ombrophylous Forest of Terra Firma. Data processing aimed to select the best regression model considering the four UPAs in the area of management. The best performance equation was chosen according to the root mean square error in percentage (RMSE%), Pearson's correlation and percent residuals graph. For the selection of the best network and its respective comparison with the best adjusted regression equation, the statistics used were: RMSE%, Pearson correlation between observed and estimated volume and bias. The best performance equation for all UPAs was the Spurr equation, which was later compared to the best RNA obtained from the data training. It was found that both methods presented acceptable adjustment and precision statistics, with potential use to estimate the volume of the species. However, RNA was slightly higher, showing greater precision in relation to the regression in the volume estimate.